 In this video, we will see other techniques for the diagnostic analytics. It is mining the data and finding the state transition between the data and how to plot it. We will see more of diagnostic analytics in coming weeks, but let us look at one such analysis in this way. Hope you remember the metal environment we showed in a week two and we said what data can be collected and how the log file might look like, right? So, we discussed that in a previous lecture. So, based on this values in the different actions stored in the MongoDB and you converted into CSV, we tried to find out how many different kind of actions a student can do in the metal learning environment. We found out that around 100 plus different actions can be done in the metal environment, different kind of actions. So, but we grouped them because 100 plus actions, 100 plus columns and you are comparing with one variable will be too complex and it would not make any inference good. So, we grouped the actions based on which part of the problem they are working on. If you remember metal, the problem map is classified into four major maps and the two calculation and evaluation, right? So, the problem map is divided into three major subtasks like is functional map, quantitative model and qualitative model. Then also it has a calculation and evaluations. Also the menu add simulator, the info and other part of tabs available for the students to use it. So, what we did, we grouped all the planning actions in the functional model as a functional model planning, all the execution or contextulations values in functional models f m e c. Similarly, for quantitative model, we name it as q m, q m p or q m e c. Similarly, for qualitative model, we have a planning and contextulations and the calculations we target as cal and evaluations we just label it as eval. It is all just a labels, you know. So, f m is one value, f m p, f from e c, the three variables similarly, q m m p, q m m e c, q m p, q m m e c, the three, nine and cal and eval. So, cal and eval, so 9, 10, 11 values, like we group all these actions into 11 different set of actions. Also, we have simulator action and info graphics. Are they using hints? Are they logging in, logging off action? So, other than 11, we have another few more actions. Let us consider these actions we have. Can we create process model of students interaction behavior in the matter? Here, we do not know what we are predicting, right? We are not predicting the students' map score or students' performance in the final test or post test. Instead of, we want to see how students is doing. What is the students interaction with the data? We might have out of say 5 students, 2 students would have done good in the post test after the intervention with the matter. 4 would have not done it. We want to say what, why happened? Why student is able to not able to score? What was the student's interaction behavior in that particular learning environment? Consider, we have a learning environment with a time stamp and all these actions are captured and we have a time series data. Let us see how to compute the transition between these variables. Before we go into looking at the methods transition, let us see the example to explain what is transition probability. Given the distribution of two states, say A and B, the two states, one is A and one is B. This is arranged in a time series data and it happened in a time series manner, right? So, first action A happened, action B A happened, B, A, A, B, something like that. They have this arranged in the action series. How many actions are A to B? Then 1, 2, 3 and 4. Some of you might know what I am doing is that I am trying to draw the state transition table or state diagram, right? If you know it is good, you can skip this video. But for others, we do not know the state transition, just trying to explain the basics. So, there are, there are A to B actions, there are 4. What is 6? 6 is indicating how many times A to other action, that is, the 6 indicates how many times A to X happen. The X can be A or B. So, if you look at A to other actions, if you look at it, this is a 1, 2, see this is a 1 transition from A, this is a second transition from A, this is a third transition from A, this is a fourth transition from A, this is a fifth, this is a sixth. A to X, it can be A or B. Out of all 6 transitions from a state A to other action, there are 4 of them to A, B only. So, 4 by 6, which means A to A will be 2 by 4 because there is 1 and 2. If there are 6 transitions from A, 4 of them are to B, which means 6 minus 4 will be 2. So, 2 by 6. Similarly, you can compute what is the transition from B to B and B to A. You can compute it, I would like you to check that one. I might be wrong, so you can please check it here. So, after you compute it, we can draw the state transition diagram. This is the state transition diagram or state diagram. Let us not go into the details of what is edges, nodes, all these things. Let us see a simple state transition diagram. This is 2 states A and B, state A, state B. From A to B, the value is 0.66. What is the probability of you will get the next action B if your current action is A, that is 4 by 6, like out of 6 actions 4 times even to B. So, high probability and 0.33 if you do an A action, there is a 33 percentage of time is possible that you will come back to the A action only. So, 0.33 is a self transition. So, there are 2 actions out of 6 came to A. So, this is basically 2 by 6. This is 4 by 6. These are the probability distribution between these 2 states. Similarly, for B that 3 by 6 and 3 by 6, so 0.5 and 0.5. Hope you understand this figure. So, this figure indicates basically the in a time series data, what is the transition between 2 actions A and B. Let us consider we have how many actions, 9 actions related to models and 2 actions calculated and other actions like a simulator and other functions. Let us consider you have like 11 actions and what will be the transition diagram based on students interaction of the system. This is the transition diagram based on students interaction of the system. One students progress and this is from this particular paper. If you want to know more about this transition, you can go and check this paper on the internet. Let us look at this transition diagram in detail. It is not indicating where the students started. That is very, very important. Start and end is not indicated. Let us consume that we start from the functional model and the color is very different. This green color indicates above 0.4. This color indicates less than 0.4, but above 0.2 like the blue color. It is less than 0.4, but above 0.2. This indicates less than 0.2. This is one students progress, one students interaction behavior in a metal system. The student who started with functional model as did the context relations, then planning a bit, but most of them he goes back to functional model. See 0.57 compared to 0.8. So, that is that compared to 0.285, the most of them is going back to the functional model. There is a self-loop and he will be like, there is a self-loop and the student is kind of going around. If we move to the functional model planning, there is high chance the student goes to qualitative model or quantitative model, both as 0.5 and 0.5. And you might see sometimes it is not adding up to one, it is because we are not showing all the values less than 0.1 in this figure 2, reduce the complexity of the figure. And there are then there is a interaction, suppose the student is goes to qualitative model, he mostly goes to context relations and planning then to quantitative model. So, that is how it indicates like this. So, there is a high transition is not no self-loop or no other actions is happening here. Might be the students reading the ins correctly or student is able to understand what have to be done next. And this indicates most students who are in evaluation go back to calculation or they go back to functional model. Also, there might be a less probability of in self-loop also less probability of less than 0.1 probability of going to other actions. So, this graph indicates how the students progress or students interaction behavior in the system. So, which way the student would have progressed? First, you would have started with functional model, then you would have spent time on functional model context relation for some time before you jump into a functional model planning. Once in the functional model planning, you would have directly jumped to qualitative quantitative model or you would have gone to quantitative model. If you do not qualitative model, you would have then qualitative model context relations planning, then you would have went to quantitative model. So, that indicates that how the student would have been progressed. And that indicates how the student would have progressed in the learning environment. So, then the student would have spent less time on evaluation or he might have moved back to functional model or he might have come back to evaluation with less probability. So, this graph indicates the students interaction behavior in the learning environment given to them. We are not able to show how much time a student spent on each of these actions or number of times the each actions occur the frequency and the time spent is not shown in this figure. If you add that information, you will get a more complete detail. We will look at it how to add those informations in the next week when we discuss about pattern mining and process mining. But this is to show you example that if you have a set of actions and you can compute simply the process model to understand what is the students progress would have been. Can you think of how this state transition plot will vary if you use interaction of more than one learner say three learners data, 10 learners data. How this transition plot will vary? Can you think of that? Just think about it if you able to think about how the plot will be then you can resume the video to continue. So, this is the process model of 6 users, 6 students using it bit lot of 2 and 4 between all these actions. What you can do is you can group the students who are able to do well in post test. Say 3 students who did good in post test after the intervention in learning environment, 3 students did not do good, you can divide them into 2 groups, you can run the process model on them like 3 students behavior analysis using the process to state transition diagram and 3 students you can plot the state transition. Then you can compare these 2 whether which one is doing good which is not doing good why the student is the group is not able to do well because they were not able to do the transition properly between the models or they stuck somewhere, they are going in a loop you can identify lot of informations without much effort of say you are not trying to identify the frequency or time just simply the sequence of actions you can complete the process model or the state transition diagram. So, in this figure it shows that it is not just a direct jump and Q and I am to QM that student would have been like directly completing QAM to QM EC instead there are lot of other loops as possible because some other students would have done that differently. So, yeah and this also tells you that if you have a proper logging mechanism and log data the N is no matter not important if N is 1 or N is 100 you can compute this kind of diagrams easily because it is all same for the computation scripting all these things. Of course, the computation time is more I am saying it will be simpler to scale if you are logging mechanism with the proper time and action IDs. So, in this video we saw an example of how to look at the learners behavior by mining their actions by plotting a state transition diagram. And next week we will discuss in detail sequential pattern mining and process mining and once we do that we will go on to a clustering mechanism that will end the diagnostic analytics part of the learning analytics course. Thank you.